Guide · Compliance
The EU AI Act and your campaign imagery
AI-generated campaign imagery needs a review of permission, audience-facing disclosure and machine-readable marking. A model release does not resolve all three.
What Article 50 requires
The AI Act’s transparency obligations (Regulation (EU) 2024/1689, Article 50) apply to “deep fakes”, AI-generated or manipulated image, audio or video content that resembles real people, places or events and could falsely appear authentic. A photoreal AI image of a person in a campaign may meet that definition, depending on context. Deployers must disclose covered content was artificially generated or manipulated, and providers of generative systems must mark outputs as AI-generated in a machine-readable format where the obligation applies. Review the Commission's current guidance and exceptions for the actual workflow.
Consent and disclosure are different layers
A common misreading: “our model consented, so no label needed.” Consent governs likeness rights, whether the person agreed to appear. Disclosure governs what the viewer is told, that AI produced the image. They are separate obligations, and a compliant campaign needs both. The practical difference is in what the label says about your brand: “AI-generated; this person doesn’t exist” reads very differently from “AI-produced imagery featuring our contracted model, who approved it and is paid for its use.”
The compliance checklist
- Inventory where AI-generated imagery of people appears: PDPs, paid social, email, OOH.
- Apply a visible disclosure where required, your ad units and product pages, not your vendor’s job.
- Check that required machine-readable marking is present and preserved. Do not assume a platform will automatically label an upload correctly.
- Keep a consent and rights record per image, the likeness layer regulators and platforms increasingly expect alongside the AI label.
FAQ
When do the EU AI Act's transparency rules take effect?
Article 50's transparency obligations apply from 2 August 2026, with scope and transitional provisions that need checking for the system involved. Consult the European Commission's current guidance.
Do AI images of real, consenting models still need disclosure?
Consent does not remove an applicable disclosure duty. Assess whether the output is a deepfake under Article 50, its publishing context and any relevant exceptions; do not assume that having a model release answers those questions.
What are the penalties for non-compliance?
Article 50 violations sit in the tier of fines up to €15 million or 3% of global annual turnover, whichever is higher.
What is machine-readable marking?
Providers of generative AI systems must mark outputs as AI-generated in a machine-readable format, for example C2PA Content Credentials, a signed manifest embedded in the image file that platforms and compliance tools can read automatically.
How Mirai handles this
Mirai's talent workflow uses real, consenting models. Draft generation is distinct from approval, commercial licensing and payment. Mirai keeps consent and licensing records and public licence verification pages today; embedded C2PA Content Credentials are in development and not yet in every licensed download.
General information, not legal advice. Consult the European Commission guidance and the current Mirai implementation status before relying on this summary.